n8n, Zapier or Make: A Decision Procedure

Every comparison of these three tools ends the same way: it depends on your needs. That is the question you arrived with, handed back to you with a table attached.

So this page is a procedure, not a comparison. It is four questions, asked in order, where each answer eliminates tools rather than describing them. Work down the list and you stop at the first question that rules something out. If you want the feature-level detail behind the rulings, our comparison of n8n, Zapier and Make for New Zealand businesses is still the right page — this one is for deciding.

The order matters. Question one is a constraint, and a constraint beats a preference every time.

Question one — must this data stay in infrastructure you control?

Zapier and Make are hosted services. Your data is processed on their systems, under their security and privacy terms. For most workflows that is entirely fine and is not worth a second thought.

It stops being fine when you have an obligation or a commitment that says otherwise. Under the Privacy Act 2020, an agency may disclose personal information to a foreign person or entity only where one of the conditions in information privacy principle 12 is met — the Privacy Commissioner’s plain-English summary of principle 12 is the readable version. Separately, and more often decisive in practice, plenty of professional engagement terms commit a firm to more than the law requires. If a workflow moves health records, client files, or anything your own client agreement says stays onshore, that is a constraint, not a preference.

If the answer is yes: n8n, self-hosted. Zapier and Make are both ruled out — not because they are insecure, but because neither can satisfy a requirement that the data never reach a third party’s infrastructure. No feature on either side changes that.

If the answer is no, carry on. Do not choose n8n for this reason if the reason does not apply; you would be buying a server to solve a problem you do not have.

Question two — does it loop, and how often does it run?

Zapier meters tasks and Make meters operations, and both are counted per unit of work rather than per workflow. A run that iterates over twenty invoice line items is not one unit — it is roughly twenty, every time it fires.

An invoice is the everyday example: a single Xero invoice carries a line-items array, and a workflow that touches each line is doing as many units of work as there are lines. So the shape that breaks the hosted pricing model is not "a busy workflow". It is a workflow that iterates, running often. Multiply your realistic iterations by your realistic run count before you decide anything, and compare it with your plan’s allowance.

If that number is large: n8n, self-hosted, where the cost is a server rather than a meter. Zapier and Make are both ruled out on economics — they will work perfectly and the invoice will be the problem.

If it is small, both remain in play, and the next two questions decide between them.

Question three — who maintains this in six months?

This is the question that is almost never asked and most often regretted.

Self-hosting n8n means somebody patches it, backs up its database, and secures it. That is ordinary work, and it is work. If no one in the business does that kind of work and you are not paying someone who does, you are buying an operational burden to save a subscription.

If nobody will own the infrastructure and question one did not force your hand: n8n is ruled out. Choosing it anyway is how a working automation quietly becomes an unpatched server.

Between the two hosted options, the tiebreak is who reads it next. A Zap is a list of steps, top to bottom, and a non-technical maintainer can usually follow one without training. A Make scenario is a canvas with routers and iterators — more capable, and genuinely harder to read cold.

Question four — does it need something the platform does not ship?

All three run code. The difference is dependencies.

Zapier and Make both let you write a code step. Neither lets you install an arbitrary library or a binary into their runtime. n8n’s code node runs in an environment you control when self-hosted, so an unusual npm package or a command-line tool is a container change rather than a dead end.

If the workflow needs a library the platform does not have, or a system tool: n8n, self-hosted. Zapier and Make are both ruled out — there is no version of either where you install your own dependency.

If it does not, and you reached this line, take whichever hosted tool won question three.

Four New Zealand scenarios, resolved

Each of these stops at a specific question. That is the point: you do not need to evaluate every feature, you need to find the first question that eliminates something.

A Hamilton trades business — invoice paid in Xero, job marked complete, team notified

Half a dozen fires a day. Popular applications on both ends; Xero has a first-party Zapier integration. The office manager will own it and does not write code or run servers.

Answer: Zapier.

• Question one: no onshore constraint. Job status and an invoice number are not client-confidential in the way a file is. Nothing eliminated.
• Question two: no iteration, low volume. The meter is irrelevant. Nothing eliminated.
• Question three: n8n ruled out. There is nobody to run the server, and no constraint from question one to justify hiring that out.
• Question four: nothing exotic. Make ruled out on the question-three tiebreak — the canvas gives this workflow nothing it needs, and it is harder for the office manager to read in twelve months’ time than a linear Zap.

A Christchurch retailer — orders split across three fulfilment paths

Moderate volume. Every order is routed by stock location, with different handling per path, and the operator is comfortable with data and wants to see where a run went wrong.

Answer: Make.

• Question one: no onshore constraint. Nothing eliminated.
• Question two: it iterates over order lines, but the volume is moderate and the operation count stays inside plan. Nothing eliminated — this is the case where you must actually do the multiplication rather than assume.
• Question three: n8n ruled out, nobody to run infrastructure. Between the hosted two, the maintainer is technical and the flow genuinely branches, which reverses the usual tiebreak.
• Question four: nothing exotic. Zapier ruled out — three-way routing with different handling per branch is exactly what a canvas is for and exactly what a linear step list makes painful.

A Wellington law firm — website enquiry opens a matter

Low volume. The firm’s engagement terms commit to client information remaining in New Zealand.

Answer: n8n, self-hosted.

• Question one: Zapier and Make both ruled out, immediately. The commitment is that the information does not reach a third party’s infrastructure, and both are third-party infrastructure. Volume, price and ease of use never get a hearing, because a constraint outranks all three.
• The firm has no one to run a server, which question three would normally treat as fatal — so the resolution is n8n run by someone who does that for a living, not a different tool. Note what happened: question one removed the option of changing the answer, so the cost lands on hosting rather than on the choice.

If you are at this point, who to talk to about building n8n workflows in New Zealand covers the three routes and what each costs you later.

An Auckland practice — reconciling line items across a large monthly batch

No confidentiality constraint that rules out hosted tools. But every run iterates over many lines, and it runs constantly.

Answer: n8n, self-hosted.

• Question one: nothing eliminated.
• Question two: Zapier and Make both ruled out. Do the multiplication and the metered cost of iteration at this volume is the whole decision. Both would work; both would cost more than the server that replaces them.
• Questions three and four are never reached. Once question two eliminates the hosted options, the remaining work is operational, not a choice.

What this procedure deliberately does not do

It does not rank the three tools, because they are not ranked — they are eliminated by circumstance, and a different business with a different constraint gets a different answer from the same four questions.

It also does not weigh features. Feature comparisons fail at exactly this task because every tool can do almost everything, so a table of ticks resolves nothing. Constraints resolve things. Meters resolve things. Who maintains it resolves things.

Two honest caveats. If your answer to question one is "I am not sure", find out before you build — it is the only question here where being wrong means starting again rather than paying more. And if you self-host, read what the community edition includes before you assume a feature is there.

If you arrived here looking for a Zapier alternative in New Zealand, the comparison of n8n, Zapier and Make names the two realistic ones and the situations where switching is the wrong move.

If you want help running the procedure against your own workflow, that is what our automation service does.

Common questions